FLIRE DSS: a Web Tool for the Management of Floods and Wildfires in Urban and Periurban Areas

Open Geosci. 2016; 8:711–727

Research Article Open Access

Giorgos Kochilakis, Dimitris Poursanidis*, Nektarios Chrysoulakis, Vassiliki Varella, Vassiliki Kotroni, Giorgos Eftychidis, Kostas Lagouvardos, Chrysoula Papathanasiou, George Karavokyros, Maria Aivazoglou, Christos Makropoulos, and Maria Mimikou FLIRE DSS: A web tool for the management of floods and wildfires in urban and periurban areas

DOI 10.1515/geo-2016-0068 are likely to flood and thus save human lives. Real-time Received Sep 25, 2015; accepted May 17, 2016 weather data from ground stations provide the necessary inputs for the calculation of the fire model in real-time, Abstract: A web-based Decision Support System, named and a high resolution weather forecast grid supports flood FLIRE DSS, for combined forest fire control and planning modeling as well as the development of “what-if” scenar- as well as flood risk management, has been developed and ios for the fire modeling. All these can be accessed byvar- is presented in this paper. State of the art tools and models ious computer sources including PC, laptop, Smartphone have been used in order to enable Civil Protection agencies and tablet either by normal network connection or by us- and local stakeholders to take advantage of the web based ing 3G and 4G cellular network. The latter is important for DSS without the need of local installation of complex soft- the accessibility of the FLIRE DSS during firefighting or res- ware and their maintenance. Civil protection agencies can cue operations during flood events. All these methods and predict the behavior of a fire event using real time data tools provide the end users with the necessary information and in such a way plan its efficient elimination. Also, dur- to design an operational plan for the elimination of the fire ing dry periods, agencies can implement “what-if” scenar- events and the efficient management of the flood events in ios for areas that are prone to fire and thus have available almost real time. Concluding, the FLIRE DSS can be easily plans for forest fire management in case such scenarios oc- transferred to other areas with similar characteristics due cur. Flood services include flood maps and flood-related to its robust architecture and its flexibility. warnings and become available to relevant authorities for visualization and further analysis on a daily basis. When Keywords: DSS System; on-line simulation; fire; flood; nat- flood warnings are issued, relevant authorities may pro- ural disaster ceed to efficient evacuation planning for the areas that

1 Introduction *Corresponding Author: Dimitris Poursanidis: Foundation for Research and Technology, Hellas, Institute of Applied and Computa- A Decision Support System is a computer-based informa- tional Mathematics, www.rslab.gr, Nikolaou Plastira 100, Vassilika Vouton, P.O. Box 1385, GR71110, Heraklion, Crete, Greece; Email: tion system which has the efficiency to support business [email protected] or organizational decision-making activities. In this envi- Giorgos Kochilakis, Nektarios Chrysoulakis: Foundation for ronment, the computer is the “silent partner” as the key Research and Technology, Hellas, Institute of Applied and Computa- factor, responsible for engagement of the computers in tional Mathematics, www.rslab.gr, Nikolaou Plastira 100, Vassilika the decision-making process. Through the computers, the Vouton, P.O. Box 1385, GR71110, Heraklion, Crete, Greece Vassiliki Varella, Giorgos Eftychidis: Algosystems S.A., Syggrou information is treated as the sixth resource besides the Avenue 206, Athens, Greece, P.O. 17672 machines, the money, the people, the materials and the Vassiliki Kotroni, Kostas Lagouvardos: Institute for Environmen- management [1]. Such systems can provide services for tal Research and Sustainable Development, National Observatory of the planning, operation and management of an organi- Athens, Athens, Greece zation in order to aid decision making. Thus, DSS intro- Chrysoula Papathanasiou, George Karavokyros, Christos duces multiple interdisciplinary aspects into the planning Makropoulos, Maria Mimikou: Department of Water Resources and Environmental Engineering, School of Civil Engineering, Na- process in complex decision environments by adding the tional Technical Univ. of Athens, 5, Iroon Politechniou St., Zografou, geospatial domain [2]. The basis of such systems is the Athens 15780, Greece Geographic Information System (GIS) that includes data Maria Aivazoglou: Department of Civil and Environmental Engi- management, graphic display, spatial modeling and spa- neering, Imperial College London, London, United Kingdom

© 2016 G. Kochilakis et al., published by De Gruyter Open. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License. 712 Ë G. Kochilakis et al. tial analysis functions. Across these common GIS decision impacts, including loss of human lives, health and qual- utilities, special features may also be included like model ity of life degradation, loss of private and public property optimization as well as statistical and spatial interaction and destruction of economic activities. At the European functions [3]. The fundamental components of a DSS are level, flood events are the most frequently reported nat- the database or the database management system (DBMS), ural disasters, affecting 25% more people than any other the model (the decision context and user criteria) or the type of natural disaster [12] while Price et al. [13, 14] recog- model based management system (MBMS) and the user in- nised that flash floods are among the costliest natural haz- terface, which is usually a standalone software on a per- ards around the globe and used lightning data to better sonal computer (PC) or nowadays, a web - based Graph- understand and predict flash floods in the Mediterranean. ical User Interface (GUI) which is accessible via any web Papagiannaki et al. [15] presented a database which in- browser from any platform [4]. The revolution in communi- cludes the weather events that have high-impact in Greece cation networks (3G, 4G cellular networks) and digital me- during 2001-2011. They found out that almost the half of dia (smartphones, tablets) has changed the original con- the weather events were the flash floods. These constitute cept of modules implemented within Commercial-Off-The- the most frequent type of the examined phenomena dur- Shelf (COTS) software or closed software which includes ing this decade. From 51 prefectures of Greece, Attica was all the components of a system. Yet the elements of a sys- among the more often influenced areas, mostly from flash tem can be distributed in components in different geo- floods. Regarding the seasonality of flash floods, autumn graphic remote locations and by implementing this archi- in Greece is actually the season with the higher frequency tecture, the use of iconic structures is apparent. This al- of rainfall and the highest accumulated rainfall, particu- lows the use of models and information from their original larly in the mainland [16–19]. As well, economic damages storage devices (hardware), reducing expenses for data resulting from wildfiresi.e. ( , the reduced ability of a burnt capture and analysis, integration and management. Web- forest to offer recreation opportunities) are also signifi- based DSS systems remain less common, especially in the cant, especially in Mediterranean regions, where their fre- field of natural disaster management, while the World quency is considerably high. This ecological degradation Wide Web technologies can provide integrated platforms becomes even more severe in the case of combined action, for the design, development, implementation and deploy- i.e. a flood event becomes more probable and more catas- ment of such Decision Support Systems. Current systems trophic when occurring in a formerly forested area that provide the end users with a broad range of capabilities has been devastated by wildfire. The occurrence and the and decision tasks, including information gathering and extent of both natural disasters strongly depend not only analysis, model building, sensitivity analysis, collabora- on the existing weather conditions in an area, but also on tion, decision implementation, spatial analysis and spa- human intervention, which is particularly pronounced in tial visualization [5]. Also, by using the web-based ap- peri-urban areas and can magnify the environmental im- proach, DSS systems become more flexible as all the com- pact. Typically, these phenomena have been investigated ponents of the system are located on the web, distributed separately, with different systems collecting information in different locations, while the calculations for the out- and modeling the resulting risk. This approach overlooks puts and the results of the models are running “on the two significant facts: fly”. Therefore, the models have been designed with opti- • The field data required in both cases are essentially mizations in order to provide real time information. Thus, the same, and hence a “collect once – use for many the information provided to the person responsible for the purposes” paradigm can be adopted resulting in in- confrontation of natural disasters will be available in a creased accuracy and economies and, manner of real-time response. DSS are popular in several • The phenomena are tightly interrelated, with forest fields like water resources management [6], renewable re- fires exacerbating the risk of flooding and preceding sources management [7], environmental management [8], floods drastically reducing the risk of fires. fire management [9, 10]) and flood management [11]. More- over, the application of a DSS is considered vital in the ar- This implies that a combined approach to manage eas of fire and flood management where the early detection flood and fire risk would achieve better, more realistic re- of the ignition of the fire or the early warning of aflood sults at a decreased cost and thus have considerable added event is crucial for the protection of human lives, proper- value beyond current practice. The fact that end-users, in ties and assets. Forest fires and flash floods has be clas- the form of emergency services (e.g. civil protection) are sified among the most destructive natural disasters, the more often the recipients of both warnings only strength- occurrence of which is related with severe socioeconomic FLIRE DSS: A web tool for the management of floods and wildfires Ë 713

Figure 1: The boundaries of Rafina catchment. ens the case for a combined risk assessment and manage- 2 Project area and datasets ment. The aim of the FLIRE DSS system is to change the The project area located in the hydrological runoff basin paradigm for the coupled, effectual and strong risk assess- of Rafina, a suburban area with an extent of2 123km , in ment and management of both flash floods and forest fires. the region of Attika, Greece. It is a typical Mediterranean This will be achieved by using state of art tools, technolo- area with mixed land uses. The area has been subjected to gies and methods, taking into account prevention, adjust- a fast and unregulated urbanization in the last decade. It is ments and interaction issues. Meanwhile this system is close to “Eleftherios Venizelos” Athens International Air- based and built on the platform of the World Wide Web. port, as well as to the A6 highway (Attiki Odos) (Figure 1). A6 highway connects the project area with Athens city, fa- voring urban sprawl [20]. Seasonal crops, sparse cultiva- 714 Ë G. Kochilakis et al. tions of vineyards and olive trees characterize the vegetation in the southern part of the project area. Low vegetation plateau can be found at the northern part of the area. Rill networks, which host riparian vegetation and pine forests, are tracked at its western and eastern part. Some small patches of pine forests can be spotted between the buildup areas, mainly in the margins of the towns. Given that its vegetation is particularly flammable, the area is prone to wildfires. The northern hills have been affected by consecutive wildfires that have burnt the natural vegetation repetitively in the last decade. Due to the lack of natural vegetation as well as the extended urbanization in the area, strong rainfalls have turned into flash floods with catastrophic reverberations to the citizens’ lives and properties, especially those properties located on the banks of the seasonal creeks. In addition, the fact that the study area is apt to forest fires and flash floods results in its grad- ual but dire ecological degradation. These areas have been traditionally used by the citizens of Athens for recreation activities including swimming, fishing, trekking and con- tact with the nature. The degradation of the environment causes significant decline to these activities. An extended network of meteorological and hydrological stations equipped with state-of-the-art sensors exists in the area. Since 2005, the Laboratory of Hydrology and Water Resources Management of the National Technical University of Athens (NTUA) operates a network of fully automatic hydrometeorological stations that covers ade- Figure 2: FLIRE DSS flowchart. quately the greater Athens area (Hydrological Observatory of Athens – HOA, [21]). Meteorological datasets are com- plemented by measurements from the dense meteorological network operated in the greater area by the National Observatory of Athens (NOA), as well as weather forecasts for the area, also provided by NOA [19]. Extended fieldwork has been done to collect field data for the preparation of the base data like the updated landcover dataset [22].

3 The FLIRE DSS System

FLIRE DSS is the web based decision support system (web- DSS) for integrated weather information management, forest fire management and floods information management. Figure 3: The architecture of the FLIRE DSS. It uses service-orientation architecture (SOA) and is based on IT sources (Information Technology) and Geoinforma- tion (GI). Fire propagation modeling and floods case sce- protected system, in which only authorized users can have narios based on weather forecasts (Figure 2) are used access for security reasons. The components of the sys- as web services. FLIRE DSS is accessible from the web tem are used as web services via a Graphical User Inter- (www.flire-dss.eu) with no prior installation of any add- face (GUI). The FLIRE web-DSS consists of three different on software for support on the browser. It is a password modules (Figure 3) under the FLIRE Server. The server uses

FLIRE DSS: A web tool for the management of floods and wildfires Ë 715

Figure 4: The Spatial Data Visualization Board.

FTP and HTTP communication protocols as well as web narios. It also includes smart alerts and scenarios for service technologies, while a GUI (Graphical Users Inter- future planning. face) has been designed and developed based on the user’s 5. FLIRE server: Unifies the aforementioned modules. requirements. The FLIRE DSS consists of five applications, 6. GUI: The user interface. described in the following sections: 1. Weather Information Management Tool (WIMT): It handles, manages and provides to the DSS the avail- 3.1 The FLIRE Server and the GUI able weather information data. As previously men- The FLIRE Server unifies the parts of the system and which tioned, these data are coming from two sources (a) can be accessed from the GUI. Users have access to the weather ground stations data & (b) weather forecast FLIRE DSS web site under a password protected system data. in order to prevent its use by non-authorized personnel. 2. Storms Early Warning System: It serves information The backbone of the FLIRE Server is a Microsoft Windows about storms in the study area. Server and the DSS tools are implemented in six different 3. Early Fire Warning System (EFiWS): It provide con- tabs: (A) Map, (B) Weather Forecast Data, (C) Weather Sta- trol of the Fire Management System and the Fire In- tions Data, (D) Fire Management System, (E) Fire Danger dex (KBDI index). Index and (F) Floodplain Data, (G) Alerts and (H) Planning 4. Flood Risk Assessment System (FLORAS): It pro- tool. vides the user with flood maps based on weather Tab A act as the Spatial Data Board (Figure 4) where forecasts and flood maps for different rainfall sce- the users can visualize data including geospatial infor- 716 Ë G. Kochilakis et al. mation, the output of the fire model, results of the flood All floodplains are available either for loading and ex- model, the weather forecast data, and the weather obser- amining in tab A or for download in KML format for further vations from the deployed surface station network as well analysis, such as time series analysis, impact on proper- as other available spatial data like satellite images pro- ties, etc. vided by Google Earth. For the needs of the monitoring of the performance of the system, statistics for each action in each tab are collected and presented in the tree menu of 3.2 Weather Information Management Tool the DSS. (WIMT). These statistics are related to the successful and failed requests of the Weather Stations Data, the Fire Manage- WIMT is implemented in two different tabs (B & C) and has ment System Data, the Available Weather Forecast Data, been designed in order to control the weather data from the Fire Danger data, the Storms detection data and the four different sources: data in the Floodplain Catalog. Tabs B and C are parts of • Weather forecast data through the National Obser- the WIMT. In tab B, users have access to the weather fore- vatory of Athens (NOA). cast data. These data are used as input to the fire models • Real-time weather observations from the meteoro- in the cases of the scenario building (“what-if” approach), logical stations operated by the National Observa- while they are also provided in XML and KML format for tory of Athens (NOA). download and further use in other applications (redistri- • Real-time weather observations from the meteoro- bution protocol). logical stations operated by the National Technical Tab C provides access to the weather stations data in University of HOA operated by NTUA. real time. These data are used in the fire modeling tools in • Storm information from National Observatory of order to examine the spread of a fire from a given point (ig- Athens (NOA). nition area) in a predefined time. Tabs D and E are parts of the EFWS. Tab D provide the control of the Fire Manage- MM5 numerical prediction model is used for the ment System. For this tab, the options are related to: weather forecast chain. It is a non-hydrostatic model, which use terrain-following coordinates [23]. It includes 1. The flash point(s) of a fire, through tabA, three nested grids (24, 8 and 2 km of spatial resolution) 2. The weather data from the station’s network through and the innermost grid cover the Athens area and the sur- tab C, rounding water bodies. MM5 runs one time per day with 3. The potential to use forecast data as input in the pro- initial time at 00:00 UTC based on the Global Forecast Sys- cess of “what-if” fire scenarios. tem (GFS) gridded analysis with a 6 hour intervals fore- The last one is necessary when users need to analyze cast [24–27]. It runs with a 72-hours forecast lead time, ex- what can be predicted to occur in the case of a fire at a given cept for Grid 3 that runs for 42- hours (initialized at 06:00 location of the study area when forecasted data imply in- UTC). creased fire risk. Tab E gives to the users the ability tovi- For the needs of the FLIRE DSS, the forecasts at 1-h sualize the Fire Danger Index values for the project area interval from the high resolution grid (2 km × 2 km) of by implementing either meteorological measurements or the requested meteorological parameters, including near weather forecasts. By using Voronoi polygons, the Fire surface temperature, relative humidity, 10-m wind speed, Danger Information is given in the spatial extend of the wind direction and rainfall are stored and used in the area. Tab F gives access to probable floodplains for the WIMT. These parameters are necessary for the estimation present and the following day. Floodplains are produced of the risk related to fires and floods. The produced forecast for three different soil moisture conditions: data are stored in text files every hour in the server ofthe 1. Probable floodplain based on rainfall forecast from National Observatory of Athens (NOA) and are retrieved by NOA (assuming normal soil moisture conditions) implementing FTP communication protocol. The forecast forecasts data are handled using the NoaForecastData tool, an inter- 2. Probable floodplain assuming wet soil moisture con- nal module operating in the DSS server that was developed ditions by FORTH for this purpose. It is executed automatically as 3. Probable floodplain assuming dry soil moisture con- a “Scheduled Task” at specific times (when forecast data ditions are available) while all parameters (urls, folder names etc.) needed by the program are stored once in NoaForecast- Data.ini file. This automatic task prepares all the available FLIRE DSS: A web tool for the management of floods and wildfires Ë 717

Figure 6: Flowchart on the use of GFMIS web service using weather stations data.

tion option” is activated and storms detection information is available, the FLIRE DSS system provides the user with an alert message, describing the area where the storms are detected, and it automatically focuses the map to that area.

Figure 5: (a) The extension of MM5 grids. The polygons define the 3.3 Early Fire Warning System (EFiWS) position of the nested grids (Grid 1 and B - intermediate and fine). (b) Athens area - Grid b. EFiWS consist of two components for fire management: • The Geographic Fire Management Information Sys- forecast data in the format that are needed for the models tem (G-FMIS): Is the model for the prediction of the and then serves these to the FLIRE DSS system. propagation of the fire; Weather forecast data are utilized in order to run • The Keetch-Byram Drought Index (KBDI): Is the in- “what-if” fire scenarios for the G-FMIS as well as on thecal- dicator on the Fire Danger risk. culation processes for the Fire Danger Index. In addition, the WIMT incorporates the weather observations of the surface weather stations that are deployed and operated 3.3.1 G-FMIS (Geographic Fire Management Information within the study area by both NOA and NTUA. These auto- System) mated weather stations provide real-time measurements of the following meteorological parameters: G-FMIS is a Fire Management model, mainly for the forest • air temperature (in ∘C) fires [28–31]. It incorporates a simulator on the forest fire, • Relative humidity (in %) based on the BEHAVE model [32]. It uses the shortest path • wind speed (in km/h) algorithm modified by Dijkstra [33] adapted to the simula- • wind direction (in ∘) tion of the forest fire spread. G-FMIS has been redesigned • rainfall (in mm) to be used as a web-service and has been incorporated in • height of station (in meters) the DSS in order to provide assessment on the fire risk and simulations on the fire propagation (Figure 6). It employs Meteorological observations are provided every 10 fuel maps, based on forest fuel mapping, as it is a precon- minutes from HOA. NTUA homogenizes, manages and dition for the forest fire propagation simulation and the fire stores data from the stations network of NOA and HOA. behavior assessment. An update fuel map has been made Data are recovered by the WIMT by applying FTP com- for the project area, based on satellite image analysis for munication tool in XML format. The data are stored and the extraction of the land cover/use [22]. managed by the weather management module of the DSS The Northern Forest Fire Laboratory (NFFL) fuel model database. Weather data from the stations are used as in- system [34] and the Prometheus [35] have been combined put in the fire models. NOA provides information about for use in the project. Prometheus is a classification sys- storms, detected in an area of 20 km around the study tem, developed for a better understanding and represen- area, every 15 minutes. When FLIRE DSS “storms detec- tation of the typical elements of the Mediterranean for- 718 Ë G. Kochilakis et al.

a request directly to GFMIS server located in a different domain. Therefore, the GFMIS component has been created on the FLIRE web server in order to process requests to GFMIS server. The GFMIS component receives requests from the FLIRE web application and forwards the requests to GFMIS server (Fig. 7). The GFMIS component processes the response of GFMIS server and creates an xml response either with an error message (describing the problem) or the results of GFMIS server.

Figure 7: Flowchart on the use of GFMIS web service using weather forecast data. 3.3.2 Fire Danger Index

The Fire Danger Index is based on the KBDI index [36]. It is est ecosystems. The fuel model, comprised by seven fuel calculated by using the station’s data or the forecast data types, is based on the height and density of fuel. These two to support “what-if” analysis based on the drought condi- have a direct influence to the intensity and propagation of tions (Figure 9). fire. Each fuel type is associated with parameters thatare The initial date for the KBDI index and the number used in the fire propagation modeling. of the days for which the information is needed are the The user has two tabs to select for the use of the G- data that the user has to provide to the system. Follow- FMIS, the standard and the advanced. The standard tab is ing, by pressing the appropriate button, the table with the the simple approach for users who do not need to parame- information appears. For each station or per weather fore- terize the system for the request time or for the simulation cast point (grid centroid), the data appear per date. For the duration and the simulation step duration, as these values KBDI index by the forecast data, a colored grid is created are fixed (120 seconds, 180 minutes and 60 minutes respec- within the project area. Four colors have been used in or- tively). The user has to provide the fire ignition point (or der to denote the values of the index: Green for values with points) in the form of click on the map at the area of in- range from 0 to 25, yellow from 25 to 100, orange from 100 terest and then has to update the system with the last 10 to 150 and red over 150. These values represent how severe minutes of data from the weather stations in the project the fire could be in case of a fire case. area. The advanced tab provides the user a set of param- eterizations of the G-FMIS, where the user can change the request time, the simulation duration and the simulation 3.4 Flood Risk Assessment System (FLORAS) step duration. This is crucial when low speed network connections are used. Also, from this control board, user can FLORAS continuously monitors hydrometeorological in- feed the model with the weather forecast data for the im- formation coming from the monitoring stations and short- plementation of fire case scenarios. These can be designed term predictions for the area. All data are assessed and for the present and the next day or by using the previous compiled by the Scenario Management System feeding day’s datasets, in order to prepare the simulation for the a catchment hydrological model, a catchment hydraulic behavior of the fire that occurred in the area and has effi- model and an urban hydraulic model, producing infor- ciently eliminate (Figure 7). mation relevant to the flood risk assessment and flood The aforementioned parameters are necessary for the alerts. The hydrological model selected for this application execution of the models. Subsequently, after the described is HEC-HMS [37], while the models selected for hydraulic steps, the user has to send the request to the G-FMIS sys- simulations are HEC-RAS [38] catchment hydraulic mod- tem and within the predefined time, the fire model runs eling and SWMM [39] for urban hydraulic modeling. All and returns a simulation on the spread of the fire for the three models used for this task are efficient, widely applied given timestep (Figure 8). and well established models. Their combined operation For each given point of the fire front, parameters such in an integrated framework has been documented in re- as the length of the flame and the step of the fire frontal search [40]. They are usually accessed over a user interface are calculated and presented. The user can have access to as stand-alone applications. For the current project, this these via the popup balloon at each point. Due to AJAX interactive approach is not an option, since all procedures, cross-domain policy, FLIRE web application cannot make from the creation of the hydrologic scenario to the produc-

FLIRE DSS: A web tool for the management of floods and wildfires Ë 719

Figure 8: Fire spread during 4 hours, by using four start points. other techniques offered by these models in order to feed them with the relevant data, run the scenarios and extract simulation results. Although it is not planned to be applied to areas other than the one specified in this Project, FLO- RAS can be adapted to other areas with modest effort by modifying the configuration files. The procedure follows the steps below (Figure 10): • Preparation of the hydrological scenario: Hourly areal rainfall forecasts in a spatial discretiza- tion of 2 × 2 km produced by NOA for the next 48 hours are downloaded and compiled. Grid points are matched to the sub-basins of the study area. The time series are disaggregated from 1 hr to 10 min time-step creating continuous time series for each catchment for the time period of three days. Figure 9: Weather forecast data downloading and processing and • Run of the hydrological model and the catchment calculation of KBDI data. hydraulic model: All files comprising a HEC-HMS project for event- tion of flood inundation maps, have to be automated and based simulations are created from scratch. Precip- itation time series are prepared using the script- processed in batch mode. FLORAS applies scripting and 720 Ë G. Kochilakis et al.

Figure 10: Flowchart on the on the production of flood results in KML.

output point of SWMM model is matched to an area- polygon of the urban zone and all these polygons are stored into a shapefile. Water depth is attributed to each polygon according to the output of SWMM simulation and different water depths are depicted with different color classes in order to be easily readable on exported maps. • Post processing and visualization of the results: The water depth at specific locations is calculated Figure 11: Flowchart on the communication of the FLIRE DSS with the from the DTM and the simulation results. The flood Floodplain repository. inundation area is calculated and stored in form of KML files. ing functionality of HEC-DSS Vue [41]. The project The resulting KML files are stored in the FLIRE files in is run by HEC-HMS in batch mode and the result- kml format, where an automatic schedule stores them in ing discharge is evaluated. Peak flow from each dis- an http server from where they can be retrieved using the charge time series representing the “worst case sce- HTTP protocol. A dictionary containing all stared flood- nario” for steady flow is identified. The maximum plain KML files is automatically updated. discharge for selected locations of the river network Each entry of this dictionary describes the output of (e.g. sub-basin outlets) is calculated and used as in- a specific hydrological/hydraulic scenario in one of the put of HEC-RAS. The resulting water elevations at available languages (English or Greek). Whenever a new the cross sections of each reach are stored in a spa- scenario runs, a new entry is inserted in the dictionary for tial data file each of the available languages and at the same time the • Run of the urban hydraulic model SWMM: output of the new scenario can be accessed through the The Storm Water Management Model (SWMM) is FLIRE web application. When these are generated, a pull used for the hydraulic simulation of the urban area. function is activated and the flood data are transferred to After the execution of HEC HMS and HEC-RAS, a the FLIRE DSS, stored internal and are prepared for the vi- simulation of SWMM is triggered. The required in- sualization upon request (Figure 11). In case no new data put is the discharge time series from the HEC-HMS are available at the time the user requests to visualize the model and rainfall input. The model provides wa- information, an appropriate message informs that no data ter depths at specific points in the urban area. Every are available up to now. In the Floodplain Data tab, the

FLIRE DSS: A web tool for the management of floods and wildfires Ë 721

Figure 12: Visualization of a flood case in the spatial data visualization board. user has access to the data in order to visualize them (Fig- ning detection in the greater study area and is useful for ure 12). the period of a couple of hours prior to a potential flood oc- Here, the user can view the spatial extend of a flood currence. The 3rd alerting level is based on forecasted data event and identify areas that are prone to flooding in the and simulations performed from the hydrological model of urban (with yellow) and peri-urban areas (with blue). In the FLIRE system and is useful for the period several hours this way, the service can support relevant authorities to up to 48 hours prior to a potential flood occurrence. Each take the necessary measures in vulnerable areas in order alerting level is described in detail below. to reduce flood impact and also assist in the identification of locations for the evacuation of the vulnerable to flood areas. 3.5.1 1st Alerting level

The 1st alerting level is provided by NTUA and is based on 3.5 Alerts real-time stage recordings from flow gauges installed in appropriate locations the study area. More specifically, four Three different smart alerting levels are provided by the flow gauges (Drafi, Spata, Rafina and Rafina2) installed system (Figure 13). The alerts are available from the tab in the area send measurements of water levels to the “Alerts”. The 1st alerting level is based on observed wa- platform. These measurements are compared against pre- ter levels at specific flow gauges in the study area andis selected threshold values for three different smart alerts: useful for the period several minutes prior to a potential yellow (lower level), amber (medium level) and red (higher flood occurrence. The 2nd alerting level is based on light- level) alert. The green color represents no Alert. When the

722 Ë G. Kochilakis et al.

Figure 13: The Smart Alerts tab with the three levels. threshold values for any of the three levels are exceeded, yellow (lower level), amber (medium level) and red (higher then the Flood Early Warning System (FEWS) of the plat- level) alert and evaluated. The green color represents no form is triggered and appropriate warnings are issued. Alert. When the threshold values for any of the three levels are exceeded, then the Flood Early Warning System (FEWS) of the platform is triggered and appropriate warn- 3.5.2 2nd Alerting level ings are issued.

nd The 2 alerting level is provided by NOA and is based on the production of smart alerts for thunderstorms with 3.6 Planning Tool a regular update approximately every 15 minutes. In particular, every 15 minutes an automated procedure is exe- This tool performs detailed cost-benefit and environmen- cuted at NOA, scanning ZEUS lightning detection network tal analyses, as well as runs of the flood model chain data for lightning occurrence within a radius of 10 kmand and eventually suggests sets of measures-interventions for 20 km around the city of Rafina. A warning is issued and flood risk management in Rafina catchment. More specifi- transferred to the FLIRE online tool when lightning is ob- cally, a list of structural and non-structural measures that served through this procedure within a radius of 10 km have been properly selected to be efficient for applica- from Rafina city (red alert), or within a radius of20km tion in the study area has been established. These mea- from Rafina city (amber alert). sures were incorporated in the hydrological model (HEC- HMS), hydraulic model (HEC-RAS) and the model chain ran for different scenarios-initial conditions in terms of rd 3.5.3 3 Alerting level fire occurrence, urban development and rainfall return period. It needs to be highlighted that, particularly for sce- The 3rd alerting level is provided by NTUA and is based narios that concern fire occurrence, a methodology de- on weather forecasts provided by NOA. The hydrological veloped during the implementation of the FLIRE Project (HEC-HMS) model of the FLIRE system automatically re- that considers floods-fires interaction has been incorpo- ceives the weather forecasts and is run for normal soil rated in the hydrological model. According to this method- moisture conditions. As mentioned in Section 3, weather ology, when a recent forest fire affects the study area, then forecast data are available every day for the current and different values need to be attributed to five properly se- the following day. Therefore, the simulations are per- lected hydrological parameters (i.e. Curve Number (CN), formed every day for the current and the following day as Initial Abstraction (IA), Standard Lag (TP), Peaking Coef- well. Simulated discharges at appropriate locations (Drafi, ficient (CP) and Muskingum K coefficient) in order toper- Spata, Rafina and Rafina2) are compared against pre- form efficient flood simulations. After running these sce- selected threshold values for three different smart alerts: narios and estimating the corresponding floodplain, the

FLIRE DSS: A web tool for the management of floods and wildfires Ë 723

Figure 14: The planning tool with the relevant information. suggested structural and non-structural measures as well rence and several combinations of these criteria. Once the as all the combinations of those measures were tested, desired scenario is selected, the platform presents: (1) a model chain runs were repeated and the corresponding map of the study area and (2) three different text boxes. The floodplains were estimated. The optimum solution (sug- map depicts the floodplain (blue color) for the selected ini- gested measure(s) for each scenario) was estimated us- tial conditions and one or more red rectangles indicate the ing socioeconomic criteria (construction and maintenance area where suggested measure(s) can be applied. The text cost and reduction of floodplain for each measure and box on the right of the map, presents the optimum selected each combination of measures) and environmental criteria solution and an Index estimated using the aforementioned (environmental footprint of each measure and each com- socioeconomic and environmental criteria, which repre- bination of measures), to which appropriate weights were sents the impact of the implementation of the particular attributed. The FLIRE platform provides access to the Plan- solution (presented as the number of stars from 1 to 5, with ning Tool for flood management by selecting the tab “Plan- increasing number of stars representing better solutions). ning Tool”. The user can choose one out of eight selected Just below this text box, other suggested solutions and pri- scenarios that correspond to different initial conditions for marily nonstructural measures that can be applied for the the model chain runs. These scenarios include the current specific scenario are suggested. Finally, a text box that de- situation in terms of urban development and fire occur- scribes in brief the functionality of the Planning Tool is rence, urban development scenarios for the next 20 and presented below the map (Figure 14). 50 years, fire scenarios in terms of fire extent, rainfall return periods that correspond to rainfalls of medium (T = 50 years) and high (T = 200 years) probability of occur-

724 Ë G. Kochilakis et al.

Table 1: Results for the calculated statistical scores for various thresholds of 24-h accumulated rainfall.

Rain Thresholds in mm 1 2.5 5 10 20 Areal Bias 0.90 0.83 0.73 0.65 0.51 POD (Probability of Detection) 0.89 0.82 0.72 0.63 0.42 FAR (False Alarm Ratio) 0.01 0.01 0.01 0.04 0.17 CSI (critical success index) 0.88 0.82 0.72 0.61 0.39

4 Discussion to analyze and plan “what-if” scenarios during dry seasons for the fire services or during wet seasons for flood services. For the case of the fire, the user has to provide The FLIRE DSS is an integrated system which combines weather forecast data to the fire model for a specific date an easily expandable web application as frontend and a (instead of the real time weather data) and then to set the backend with several web services, large volume of real ignition point of the fire. These could be valuable in areas and non-real time data from different providers and appli- that are prone to fire due to flammable vegetation. By send- cations for data processing and map coverages tiles cre- ing the data to the model, the results appear on screen in ation. The frontend is a Javascript web application that few seconds. The results of the model can be used for anal- is based on Google Maps API (to display static maps and ysis in an efficient planning in this area. Floods are cal- dynamic map information) and uses Ajax requests to re- culated daily by using the forecast data and datasets are trieve data from the system web services. The backend available for visualization in the system. FLIRE DSS pro- data are organized in xml and kml data files and the ap- vides easy access to the system’s components via a web- plications are written in Visual Basic. The developed web site that hosts the platform instead of a classic desktop ap- services provide the web application (or any other autho- plication. Moreover, all the components of the system are rized user or application) with both real and non-real time geographically isolated, providing the system the desired data and data dictionaries. FLIRE DSS has been intensively distributed architecture. The fire model has been designed checked for its performance and reliability by accessing it as a web service and has as a backbone, the well-known from different technological sources. It has been accessed BEHAVE model [32] which is widely used in the era of wild- and tested by office computers and laptops with wired and fire management. The well-established models HEC-HMS, wireless network connections and different operation soft- HEC-RAS and SWMM have been selected for flood model- ware (Microsoft Windows, Linux, Apple Mac OS), tablet ing. The hydrological model HEC-HMS and the catchment and smartphones with 2G/3G/4G network and various op- hydraulic model HEC-RAS have already been successfully eration systems (Apple iOS 8, Android) from remote areas. applied several times in the study area in event-based Meanwhile, the system has been demonstrated to the Civil mode (e.g. [42, 43]). Efficient calibration of both models Protection in order to have an interplay for improvements was achieved through the cross-validation of simulated re- on the design and the presented tools. sults with observed datasets, when available. The perfor- The FLIRE DSS is a promising tool for the fire brigade mance of the FLIRE DSS regarding the ability of the meteo- agencies, the Civil Protection and the local stakeholders. rological model to quantitatively forecast rainfall has been Usually, these agencies have the knowledge and the funds evaluated. The verification covered the period from March in order to operate and manage their own IT systems. The 2013 up to December 2014, a period that comprises of 37 FLIRE DSS provides natural disaster management depart- rain episodes, with at least one station recording more ments with the advantages of the GIS abilities without the than 20 mm of rain within 24 hours. For the verification challenge of the installation of complicated and expensive period, 44 rain gauges were selected, operated by the Na- software. The platform provides the users with real time in- tional Observatory of Athens and the National Technical formation on the current weather conditions in the area as University of Athens. Following the methodology widely well as high resolution imageries which support the iden- accepted for evaluation of precipitation, a contingency ta- tification of the shortest paths to reach the areas offireor ble (yes/no for observed/modelled rain) was constructed flood and other important information for the surrounding for the totality of the 37 episodes and several statistical terrain. Important component of the FLIRE DSS is the po- scores were calculated. The calculation of the statistical tential of the users to use weather forecast data in order scores (Table 1) revealed a decreasing trend of the Prob- FLIRE DSS: A web tool for the management of floods and wildfires Ë 725

Table 2: Results for the calculated statistical scores for various ranges of 24-h accumulated rainfall.

Rain Ranges in mm 0.2–2.5 2.5–5 5.–10. 10.–20. >20. QB 0.28 −0.52 −1.65 −3.59 −14.36 MAE 0.66 1.02 3.05 5.97 15.95 ability of Detection (POD) with increasing rain threshold, tection and to the stakeholders on the field. By using, it with a POD of 0.42 for the highest precipitation amounts. will bring onboard the fire behavior potential in order to This result is in agreement with the verification results of plan the efficient elimination, satellite images for naviga- similar activities of high-resolution rain forecasts in the tion and spatial information related to the flooded areas. Mediterranean area [44–46]. On the other hand, the cal- FLIRE DSS is applied in East Attika, but due to its architec- culated False Alarm Ratio (FAR) was very low (lower than ture and flexibility, can be transferred to other areas or toa 0.17) for all rain thresholds, indicating that the model has broader area (region, country). The only requirements are no tendency to provide false alarms. Verification of the the creation of a fuel map and a landuse/landcover map quantity of forecasted rain against observations (calcula- for the desired area by using accurate and updated Earth tion of mean error and mean absolute errors) is provided Observation data as well as information from a network of in Table 2. ground weather stations. Weather forecast data can be cal- The model underpredicts the amounts of rain for all culated by using the same technique. Minor hardware up- ranges (negative QB) with the exception of the first range grade will be necessary for the storage of bigger datasets of 0.2–2.5 mm. The greatest errors are obtained for the large and the control of the traffic due to the accessibility ofmore amounts of rainfall (for > 20 mm the mean value of QB is users in cases of its transfer in a region level or country −14.36 mm). The same comments apply for the MAE values level. that range from 0.66 mm for the lowest range, 3 mm for the medium precipitation amounts (5-10 mm range) and an Acknowledgement: This research has been carried error value of ~15 mm for the high precipitation amounts. out during the implementation of the Project FLIRE The numbers reported in Table 2 are better that the values “FLoods and fIre Risk assessment and managEment” reported in previous studies over Greece [26, 44]. It should (LIFE11ENV/GR/975). The Project aims to develop a warn- be noted however that these previous studies refer only to ing system for floods and fire risk management and is summer period cases (when rain forecast is in general a co-financed by European Commission General Directorate more demanding task) while the analysis performed in the for the Environment, LIFE financial instrument with 50%. frame of this work spans on cases throughout the year.

References 5 Conclusions [1] Simonovic, S.P. 1996. Decision Support Systems For Sustain- The performance of the FLIRE DSS (WIMT, G-FMIS, and able Management Of Water Resources 2. Case Studies, Water International, 21(4), 233-244. Fire Danger Index) has been also evaluated. A series of [2] Chrysoulakis, N., Mitraka, Z., Diamantakis, E., González, A., 250 ignition point scenarios using weather data have been Castro, E. A., San Josë, R. and I. Blecic. 2010. Accounting for ur- used in order to evaluate the response time of the model. ban metabolism in urban planning. The case of BRIDGE. Proc. The fire model has a response in about 27 seconds (mean 10th International Conference on Design & Decision Support value) after a cold start with a maximum at 61 seconds and Systems in Architecture and Urban Planning, organized by the a minimum at 23 seconds. In some cases, the system needs Technical University of Einhoven, in Eindhoven, The Nether- lands (July 19-22). more that 35 seconds. Similar behavior has been noted by [3] Nyerged, T. 2010. Regional and urban GIS: a decision support using the weather forecast data. Flood data are transferred approach, A Division of Guilford Publications, Inc, New York, to the FLIRE DSS system when they are available and there USA. is no time lag for the visualization. The ultimate goal of [4] Böhner, C. 2006. Decision-support systems for sustainable ur- the FLIRE DSS is the migration to a mobile platform, in or- ban Planning. International Journal of Environmental Technol- ogy and Management, 6, 193–205 der to become available to the firefighters, to the civil pro- 726 Ë G. Kochilakis et al.

[5] Bhargava, K.H., Power, J.D. and Sun D. 2007. Progress in Web- 1363–1367. based decision support technologies. Decision Support Sys- [18] Kotroni V., Lagouvardos K., Kallos G. and Ziakopoulos D. 1999. tems 43(4),1083-1095. Severe flooding over central and southern Greece associated [6] Koutsoyiannis, D., Karavokiros G., Efstratiadis A., Mamassis N., with pre-cold frontal orographic lifting. Quarterly Journal of Koukouvinos A. and A. Christofides. 2003. A decision support Royal Meteorological Society, 125, 967-991. system for the management of the water resource system of [19] Lagouvardos K., Kotroni V., Bezes A., Koletsis I., Kopania T., Athens, Physics and Chemistry of the Earth, 28 (14-15), 599– Lykoudis S., Mazarakis N., Papagiannaki K. and Vougioukas S. 609. 2016. The Automatic Weather Stations Network of the National [7] Giupponi, C. and Sgobbi, A. 2013. Decision Support Systems for Observatory of Athens: Operation and Database. Geoscience Water Resources Management in Developing Countries: Learn- Data Journal (in press). ing from Experiences in Africa. Water 5, 798-818. [20] Chrysoulakis, N., Mitraka, Z., Stathopoulou, M. and Cartalis, C. [8] McIntosh, B.S., Ascough II, J.C., Twery, M., Chew, J., Elmahdi, 2013. A comparative analysis of the urban web of the greater A., Haase, D., Harou, J.J., Hepting, D., Cuddy, S., Jakeman, A.J., Athens agglomeration for the last 20 years period on the ba- Chen, S., Kassahun, A., Lautenbach, S., Matthews, K., Mer- sis of Landsat imagery. Fresenius Environmental Bulletin, 22, ritt, W., Quinn, N.W.T., Rodriguez-Roda, I., Sieber, S., Stavenga, 2139–2144. M., Sulis, A. Ticehurst, J., Volk, M., Wrobel, M., van Delden, [21] Papathanasiou, C., Makropoulos, C., Baltas, E. and Mimikou, M. H., El-Sawah S., Rizzoli A., and Voinov, A. 2011. Environmen- 2013. The Hydrological Observatory of Athens: a state-of-the-art tal decision support systems (EDSS) development – challenges network for the assessment of the hydrometeorological regime and best practices. Environmental Modelling and Software, 26, of Attica. Proc. 13th International Conference on Environmen- 1389–1402 tal Science and Technology, 5th-7th September 2013, Athens, [9] Kalabokidis, K., Athanasis N., Gagliardi F., Karayiannis F., Greece. Palaiologou P., Parastatidis S. and Vasilakos C. 2013. Virtual [22] Poursanidis D, Mitraka Z andChrysoulakis N. 2015. Landsat 8 vs. Fire: A web-based GIS platform for forest fire control. Ecologi- Landsat 5: A comparison based on urban and peri-urban land cal Informatics 16, 62-69. cover mapping. International Journal of Applied Earth Observa- [10] Noonan-Wright, E.K., Opperman, T.S., Finney, A.M., Zimmer- tion and Geoinformation 35, 259–269. man, G., Seli, R., Elenz, L, Calkin, D. and Fiedler J.. 2011. Devel- [23] Dudhia, J., (1993). A non-hydrostatic version of the Penn State– oping the US Wildland Fire Decision Support System. Journal of NCAR Mesoscale Model: Validation tests and simulation of Combustion. Article ID 168473. doi:10.1155/2011/168473. an Atlantic cyclone and cold front, American Meteorological [11] Honghai Qi and Altinakar M.S. 2011. A GIS-based decision sup- Society, 121, 1493–1513, DOI: http://dx.doi.org/10.1175/1520- port system for integrated flood management under uncertainty 0493(1993)121<1493:ANVOTP> 2.0. CO;2. with two dimensional numerical simulations. Environmental [24] Kotroni V. and Lagouvardos K. 2001. Precipitation forecast Modelling & Software 26(6),817–82. skill of different convective parameterization and microphysical [12] Prevention Web 2011. Disaster statistics from the Prevention schemes: application for the cold season over Greece. Geophys- Web http://www.preventionweb.net/english/countries/statist ical Research Letters, 28(10), 1977–1980. ics/index_region.php?rid=3. [25] Kotroni V., Lagouvardos K. and Retalis A. 2011. The heat wave [13] Price C., Yair Y., Mugnai A., Lagouvardos K., Llasat M. C., of June 2007 in Athens, Greece - Part 2: Modelling study and Michaelides S., Dayan U., Dietrich S., Di Paola F., Galanti E., sensitivity experiments. Atmospheric Research, 100, 1–11. Garrote L., Harats N., Katsanos D., Kohn M., Kotroni V., Llasat- [26] Kotroni V. and Lagouvardos K. 2004 Evaluation of MM5 High- Botija, M. Lynn B., Mediero L., Morin E., Nicolaides K., Rozalis Resolution Real-Time Forecasts over the Urban Area of Athens, S., Savvidou K. and Ziv B. 2011b: Using lightning data to better Greece. Journal of Applied Meteorology, 43, 1666–1678. understand and predict flash floods in the Mediterranean. Sur- [27] Akylas E., Kotroni V. and Lagouvardos K. 2007. Sensitivity of veys in Geophysics 32(6), pp. 733–751. high resolution operational weather forecasts to the choice of [14] Price C., Yair Y., Mugnai A., Lagouvardos K., Carmen Llasat M., the planetary boundary layer scheme, Atmospheric Research, Michaelides S., Dietrich S., Galanti E., Garrote L., Katsanos D., 84, 49–57. Kotroni V., Llasat-Botija M., Mediero L., Morin E., Nicolaides K., [28] Eftichidis G., Margaritis E., Sfiris A. and Varela V. 1998 Fire man- Savvidou K. and and Ziv B. 2011a. The FLASH Project: Using light- agement information systems: FMIS. Proc. III Int. Conference on ning data to better understand and predict flash floods. Environ- Forest Fire Research / 14th Conference on Fire and Forest Mete- mental Science & Policy, 14, 898-911. orology (Luso, 16/20 November 1998), Vol. II: 2641–2642. [15] Papagiannaki K., Lagouvardos K. and Kotroni V. 2013. A [29] Andrews, P. L. and C. H. Chase. 1989. BEHAVE: fire behavior pre- database of high-impact weather events in Greece: a descrip- diction and fuel modeling system - BURN subsystem, part 2. tive impact analysis for the period 2001–2011. Natural Hazards General Technical Report INT-260. Ogden, UT: U.S. Department and Earth System Sciences, 13, 727–736. of Agriculture, Forest Service, Intermountain Research Station. [16] Jansa, A., Alpert, P., Arbogast, P., Buzzi, A., Ivancan-Picek, B., [30] Andrews, P. L. 1986. BEHAVE: Fire behavior prediction and fuel Kotroni, V., Llasat, M. C., Ramis, C., Richard, E., Romero, R. and modeling system — BURN Subsystem, Part 1, Gen. Tech. Rep. Speranza, A.: MEDEX: a general overview. Natural Hazards and INT-194. Ogden, UT: U.S. Department of Agriculture, Forest Ser- Earth System Sciences, 14 (8), 1965-1984, doi:10.5194/nhess- vice, Intermountain Forest and Range Experiment Station. 14-1965-2014, 2014 [31] Burgan R.E. 1988) 1988 revisions to the 1978 National Fire- [17] Lagouvardos K. and Kotroni V. 2007. TRMM and lightning ob- Danger Rating System. Res. Pap. SE-273. Asheville, NC: U.S. De- servations of a low-pressure system over the Eastern Mediter- partment of Agriculture, Forest Service, Southeastern Forest Ex- ranean. Bulletin of the American Meteorological Society, 88, periment Station. FLIRE DSS: A web tool for the management of floods and wildfires Ë 727

[32] Burgan, R. E. and R. C. Rothermel. 1984. BEHAVE: fire behavior [41] Hydrologic Engineering Center. 2009. HEC-DSSVue. HEC Data prediction and fuel modeling system - FUEL subsystem. General Storage System Visual Utility Engine. User’s Manual v. 2.0. US Technical Report INT-167. Ogden, UT: U.S. Department of Agri- Army Corps of Engineers. culture, Forest Service, Intermountain Forest and Range Experi- [42] Papathanasiou C., Alonistioti D., Kasella A., Makropoulos C. and ment Station. Mimikou M.2012. The impact of forest fires on the vulnerability [33] Dijkstra, E. W. 1959. A note on two problems in connection with of peri-urban catchments to flood events (The case of the East- graphs. Numerische Mathematik 1, 269–271. ern Attica region). Special Issue of the Global NEST Journal on [34] Albini, F. 1976. Estimating Wildfire Behavior and Effects. USDA Hydrology and Water Resources, 14(3), 294-302. Forest Service. General Technical Report INT-30. 1976. [43] Papathanasiou, C., Safiolea, E., Makropoulos, C. and Mimikou, [35] Prometheus, S.V. Project, 1999. Management techniques for M. 2009. The FLADAR Project and its contribution to the imple- optimization of suppression and minimization of wildfire ef- mentation of the EU Flood Directive 2007/60. Proc. 11th Inter- fects. System Validation. European Commission. Contract num- national Conference on Environmental Science and Technology, ber ENV4-CT98-0716. 3-5 September 2009, Chania, Crete, Greece. [36] Keetch, John J & Byram, George. 1968. A drought index for forest [44] Mazarakis, N., Kotroni V., Lagouvardos K. and Argiriou A.A. fire control. Res. Paper SE-38. Asheville, NC: U.S. Department 2009. The sensitivity of numerical forecasts to convective pa- of Agriculture, Forest Service, Southeastern Forest Experiment rameterization during the warm period and the use of lighting Station. data as an indicator for convective occurrence. Atmospheric Re- [37] Hydrologic Engineering Center. 2010a. Hydrologic Modeling search 94, 704-714. System HEC-HMS, User’s Manual v.3.5. US Army Corps of Engi- [45] Sindosi O. A., Bartzokas A., Kotroni V. and Lagouvardos K. 2012. neers Verification of precipitation forecasts of MM5 model over Epirus, [38] Hydrologic Engineering Center. 2010b. HEC-RAS, River Analysis NW Greece, for various convective parameterization schemes, System, User’s Manual v. 4.1. US Army Corps of Engineers. Natural Hazards and Earth System Sciences, 12, 1393–1405. [39] Rossman, L. A., 2010. Storm Water Management Model User’s [46] Oberto E., Milelli M., Pasi F. and Gozzini B., 2012. Intercompar- Manual, EPA/600/R-05/040, U.S. Environmental Protection ison of two meteorological limited area models for quantitative Agency, Cincinnati, OH. precipitation forecast verification. Natural Hazards and Earth [40] Robayo, O. 2005. Map to Map: Converting a NEXRAD Rainfall System Sciences 12, 591-606. DOI: 10.5194/nhess-12-591-2012. Map into a Flood Inundation Map. The University of Texas at Austin, Dissertation.